Copy stable/10@r272459 to releng/10.1 as part of
the 10.1-RELEASE process.

Approved by: re (implicit)
Sponsored by: The FreeBSD Foundation

Copy head (r256279) to stable/10 as part of the 10.0-RELEASE cycle.

Approved by: re (implicit)
Sponsored by: The FreeBSD Foundation

Revert (once again, and hopefully for the last time) to flock(2) locks.
The problem with fcntl(2) locks is that they are not inherited by child
processes. This breaks pidfile(3), where the common idiom is to open
and lock the PID file before daemonizing.

Diff reduction against Varnish, including one important fix: use a shared
lock if the file is opened with O_RDONLY.

Reimplement flopen(3) using fcntl(2) locks instead of flock(2) locks.

Back out previous commit until I figure out why my regression test fails.

Approved by: re (kensmith)

Use fcntl(2)-style locks instead of less-portable flock(2)-style locks.

Approved by: re (kensmith)

Nit: avoid shadowing truncate(2) with a local variable.

Fix stupid braino in previous commit.

If (flags & O_TRUNC), don't truncate the file until we've successfully
locked it.

MFC after: 3 weeks

Well gag me with a spoon... I'm so used to working at high WARNS levels
that I make stupid fundamental mistakes like this when I don't.

Remove superfluous unexpanded RCS tag.

DTRT when O_NONBLOCK is specified.

MFC after: 3 weeks

I'm tired of seeing this done incorrectly and non-portably, so add a
flopen(3) function which reliably opens and locks a file.

MFC after: 3 weeks